How to use 'future' data to predict 'future' target with VARMAX model?

[Xudong Fan]

Hi, 

I am working on a multivariate time-series data with one predict target problem. I am using VAR model, so the prediction problem is like 

y_t = sum(a*y) + sum(b*x) + error. 

the y contains the time from t-p to t-1. But the x contains from t-p to t+k, i.e. the x contains the future data for the prediction of y_t. 

Is there a workaround in statsmodel that could handle this? I am thinking to use a LinearRegression model from other libraries such as sklearn. But I am not sure how to capture the uncertainty of the errors. 

Any suggestions?

Thanks so much!

Xudong